xref: /linux/net/openvswitch/conntrack.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2015 Nicira, Inc.
4  */
5 
6 #include <linux/module.h>
7 #include <linux/openvswitch.h>
8 #include <linux/tcp.h>
9 #include <linux/udp.h>
10 #include <linux/sctp.h>
11 #include <linux/static_key.h>
12 #include <linux/string_helpers.h>
13 #include <net/ip.h>
14 #include <net/genetlink.h>
15 #include <net/netfilter/nf_conntrack_core.h>
16 #include <net/netfilter/nf_conntrack_count.h>
17 #include <net/netfilter/nf_conntrack_helper.h>
18 #include <net/netfilter/nf_conntrack_labels.h>
19 #include <net/netfilter/nf_conntrack_seqadj.h>
20 #include <net/netfilter/nf_conntrack_timeout.h>
21 #include <net/netfilter/nf_conntrack_zones.h>
22 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
23 #include <net/ipv6_frag.h>
24 
25 #if IS_ENABLED(CONFIG_NF_NAT)
26 #include <net/netfilter/nf_nat.h>
27 #endif
28 
29 #include <net/netfilter/nf_conntrack_act_ct.h>
30 
31 #include "datapath.h"
32 #include "drop.h"
33 #include "conntrack.h"
34 #include "flow.h"
35 #include "flow_netlink.h"
36 
37 struct ovs_ct_len_tbl {
38 	int maxlen;
39 	int minlen;
40 };
41 
42 /* Metadata mark for masked write to conntrack mark */
43 struct md_mark {
44 	u32 value;
45 	u32 mask;
46 };
47 
48 /* Metadata label for masked write to conntrack label. */
49 struct md_labels {
50 	struct ovs_key_ct_labels value;
51 	struct ovs_key_ct_labels mask;
52 };
53 
54 enum ovs_ct_nat {
55 	OVS_CT_NAT = 1 << 0,     /* NAT for committed connections only. */
56 	OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
57 	OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
58 };
59 
60 /* Conntrack action context for execution. */
61 struct ovs_conntrack_info {
62 	struct nf_conntrack_helper *helper;
63 	struct nf_conntrack_zone zone;
64 	struct nf_conn *ct;
65 	u8 commit : 1;
66 	u8 nat : 3;                 /* enum ovs_ct_nat */
67 	u8 force : 1;
68 	u8 have_eventmask : 1;
69 	u16 family;
70 	u32 eventmask;              /* Mask of 1 << IPCT_*. */
71 	struct md_mark mark;
72 	struct md_labels labels;
73 	char timeout[CTNL_TIMEOUT_NAME_MAX];
74 	struct nf_ct_timeout *nf_ct_timeout;
75 #if IS_ENABLED(CONFIG_NF_NAT)
76 	struct nf_nat_range2 range;  /* Only present for SRC NAT and DST NAT. */
77 #endif
78 };
79 
80 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
81 #define OVS_CT_LIMIT_UNLIMITED	0
82 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
83 #define CT_LIMIT_HASH_BUCKETS 512
84 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
85 
86 struct ovs_ct_limit {
87 	/* Elements in ovs_ct_limit_info->limits hash table */
88 	struct hlist_node hlist_node;
89 	struct rcu_head rcu;
90 	u16 zone;
91 	u32 limit;
92 };
93 
94 struct ovs_ct_limit_info {
95 	u32 default_limit;
96 	struct hlist_head *limits;
97 	struct nf_conncount_data *data;
98 };
99 
100 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
101 	[OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
102 };
103 #endif
104 
105 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
106 
107 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
108 
109 static u16 key_to_nfproto(const struct sw_flow_key *key)
110 {
111 	switch (ntohs(key->eth.type)) {
112 	case ETH_P_IP:
113 		return NFPROTO_IPV4;
114 	case ETH_P_IPV6:
115 		return NFPROTO_IPV6;
116 	default:
117 		return NFPROTO_UNSPEC;
118 	}
119 }
120 
121 /* Map SKB connection state into the values used by flow definition. */
122 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
123 {
124 	u8 ct_state = OVS_CS_F_TRACKED;
125 
126 	switch (ctinfo) {
127 	case IP_CT_ESTABLISHED_REPLY:
128 	case IP_CT_RELATED_REPLY:
129 		ct_state |= OVS_CS_F_REPLY_DIR;
130 		break;
131 	default:
132 		break;
133 	}
134 
135 	switch (ctinfo) {
136 	case IP_CT_ESTABLISHED:
137 	case IP_CT_ESTABLISHED_REPLY:
138 		ct_state |= OVS_CS_F_ESTABLISHED;
139 		break;
140 	case IP_CT_RELATED:
141 	case IP_CT_RELATED_REPLY:
142 		ct_state |= OVS_CS_F_RELATED;
143 		break;
144 	case IP_CT_NEW:
145 		ct_state |= OVS_CS_F_NEW;
146 		break;
147 	default:
148 		break;
149 	}
150 
151 	return ct_state;
152 }
153 
154 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
155 {
156 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
157 	return ct ? READ_ONCE(ct->mark) : 0;
158 #else
159 	return 0;
160 #endif
161 }
162 
163 /* Guard against conntrack labels max size shrinking below 128 bits. */
164 #if NF_CT_LABELS_MAX_SIZE < 16
165 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
166 #endif
167 
168 static void ovs_ct_get_labels(const struct nf_conn *ct,
169 			      struct ovs_key_ct_labels *labels)
170 {
171 	struct nf_conn_labels *cl = NULL;
172 
173 	if (ct) {
174 		if (ct->master && !nf_ct_is_confirmed(ct))
175 			ct = ct->master;
176 		cl = nf_ct_labels_find(ct);
177 	}
178 	if (cl)
179 		memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
180 	else
181 		memset(labels, 0, OVS_CT_LABELS_LEN);
182 }
183 
184 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
185 					const struct nf_conntrack_tuple *orig,
186 					u8 icmp_proto)
187 {
188 	key->ct_orig_proto = orig->dst.protonum;
189 	if (orig->dst.protonum == icmp_proto) {
190 		key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
191 		key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
192 	} else {
193 		key->ct.orig_tp.src = orig->src.u.all;
194 		key->ct.orig_tp.dst = orig->dst.u.all;
195 	}
196 }
197 
198 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
199 				const struct nf_conntrack_zone *zone,
200 				const struct nf_conn *ct)
201 {
202 	key->ct_state = state;
203 	key->ct_zone = zone->id;
204 	key->ct.mark = ovs_ct_get_mark(ct);
205 	ovs_ct_get_labels(ct, &key->ct.labels);
206 
207 	if (ct) {
208 		const struct nf_conntrack_tuple *orig;
209 
210 		/* Use the master if we have one. */
211 		if (ct->master)
212 			ct = ct->master;
213 		orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
214 
215 		/* IP version must match with the master connection. */
216 		if (key->eth.type == htons(ETH_P_IP) &&
217 		    nf_ct_l3num(ct) == NFPROTO_IPV4) {
218 			key->ipv4.ct_orig.src = orig->src.u3.ip;
219 			key->ipv4.ct_orig.dst = orig->dst.u3.ip;
220 			__ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
221 			return;
222 		} else if (key->eth.type == htons(ETH_P_IPV6) &&
223 			   !sw_flow_key_is_nd(key) &&
224 			   nf_ct_l3num(ct) == NFPROTO_IPV6) {
225 			key->ipv6.ct_orig.src = orig->src.u3.in6;
226 			key->ipv6.ct_orig.dst = orig->dst.u3.in6;
227 			__ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
228 			return;
229 		}
230 	}
231 	/* Clear 'ct_orig_proto' to mark the non-existence of conntrack
232 	 * original direction key fields.
233 	 */
234 	key->ct_orig_proto = 0;
235 }
236 
237 /* Update 'key' based on skb->_nfct.  If 'post_ct' is true, then OVS has
238  * previously sent the packet to conntrack via the ct action.  If
239  * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
240  * initialized from the connection status.
241  */
242 static void ovs_ct_update_key(const struct sk_buff *skb,
243 			      const struct ovs_conntrack_info *info,
244 			      struct sw_flow_key *key, bool post_ct,
245 			      bool keep_nat_flags)
246 {
247 	const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
248 	enum ip_conntrack_info ctinfo;
249 	struct nf_conn *ct;
250 	u8 state = 0;
251 
252 	ct = nf_ct_get(skb, &ctinfo);
253 	if (ct) {
254 		state = ovs_ct_get_state(ctinfo);
255 		/* All unconfirmed entries are NEW connections. */
256 		if (!nf_ct_is_confirmed(ct))
257 			state |= OVS_CS_F_NEW;
258 		/* OVS persists the related flag for the duration of the
259 		 * connection.
260 		 */
261 		if (ct->master)
262 			state |= OVS_CS_F_RELATED;
263 		if (keep_nat_flags) {
264 			state |= key->ct_state & OVS_CS_F_NAT_MASK;
265 		} else {
266 			if (ct->status & IPS_SRC_NAT)
267 				state |= OVS_CS_F_SRC_NAT;
268 			if (ct->status & IPS_DST_NAT)
269 				state |= OVS_CS_F_DST_NAT;
270 		}
271 		zone = nf_ct_zone(ct);
272 	} else if (post_ct) {
273 		state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
274 		if (info)
275 			zone = &info->zone;
276 	}
277 	__ovs_ct_update_key(key, state, zone, ct);
278 }
279 
280 /* This is called to initialize CT key fields possibly coming in from the local
281  * stack.
282  */
283 void ovs_ct_fill_key(const struct sk_buff *skb,
284 		     struct sw_flow_key *key,
285 		     bool post_ct)
286 {
287 	ovs_ct_update_key(skb, NULL, key, post_ct, false);
288 }
289 
290 int ovs_ct_put_key(const struct sw_flow_key *swkey,
291 		   const struct sw_flow_key *output, struct sk_buff *skb)
292 {
293 	if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
294 		return -EMSGSIZE;
295 
296 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
297 	    nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
298 		return -EMSGSIZE;
299 
300 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
301 	    nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
302 		return -EMSGSIZE;
303 
304 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
305 	    nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
306 		    &output->ct.labels))
307 		return -EMSGSIZE;
308 
309 	if (swkey->ct_orig_proto) {
310 		if (swkey->eth.type == htons(ETH_P_IP)) {
311 			struct ovs_key_ct_tuple_ipv4 orig;
312 
313 			memset(&orig, 0, sizeof(orig));
314 			orig.ipv4_src = output->ipv4.ct_orig.src;
315 			orig.ipv4_dst = output->ipv4.ct_orig.dst;
316 			orig.src_port = output->ct.orig_tp.src;
317 			orig.dst_port = output->ct.orig_tp.dst;
318 			orig.ipv4_proto = output->ct_orig_proto;
319 
320 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
321 				    sizeof(orig), &orig))
322 				return -EMSGSIZE;
323 		} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
324 			struct ovs_key_ct_tuple_ipv6 orig;
325 
326 			memset(&orig, 0, sizeof(orig));
327 			memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
328 			       sizeof(orig.ipv6_src));
329 			memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
330 			       sizeof(orig.ipv6_dst));
331 			orig.src_port = output->ct.orig_tp.src;
332 			orig.dst_port = output->ct.orig_tp.dst;
333 			orig.ipv6_proto = output->ct_orig_proto;
334 
335 			if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
336 				    sizeof(orig), &orig))
337 				return -EMSGSIZE;
338 		}
339 	}
340 
341 	return 0;
342 }
343 
344 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
345 			   u32 ct_mark, u32 mask)
346 {
347 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
348 	u32 new_mark;
349 
350 	new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
351 	if (READ_ONCE(ct->mark) != new_mark) {
352 		WRITE_ONCE(ct->mark, new_mark);
353 		if (nf_ct_is_confirmed(ct))
354 			nf_conntrack_event_cache(IPCT_MARK, ct);
355 		key->ct.mark = new_mark;
356 	}
357 
358 	return 0;
359 #else
360 	return -ENOTSUPP;
361 #endif
362 }
363 
364 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
365 {
366 	struct nf_conn_labels *cl;
367 
368 	cl = nf_ct_labels_find(ct);
369 	if (!cl) {
370 		nf_ct_labels_ext_add(ct);
371 		cl = nf_ct_labels_find(ct);
372 	}
373 
374 	return cl;
375 }
376 
377 /* Initialize labels for a new, yet to be committed conntrack entry.  Note that
378  * since the new connection is not yet confirmed, and thus no-one else has
379  * access to it's labels, we simply write them over.
380  */
381 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
382 			      const struct ovs_key_ct_labels *labels,
383 			      const struct ovs_key_ct_labels *mask)
384 {
385 	struct nf_conn_labels *cl, *master_cl;
386 	bool have_mask = labels_nonzero(mask);
387 
388 	/* Inherit master's labels to the related connection? */
389 	master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
390 
391 	if (!master_cl && !have_mask)
392 		return 0;   /* Nothing to do. */
393 
394 	cl = ovs_ct_get_conn_labels(ct);
395 	if (!cl)
396 		return -ENOSPC;
397 
398 	/* Inherit the master's labels, if any. */
399 	if (master_cl)
400 		*cl = *master_cl;
401 
402 	if (have_mask) {
403 		u32 *dst = (u32 *)cl->bits;
404 		int i;
405 
406 		for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
407 			dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
408 				(labels->ct_labels_32[i]
409 				 & mask->ct_labels_32[i]);
410 	}
411 
412 	/* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
413 	 * IPCT_LABEL bit is set in the event cache.
414 	 */
415 	nf_conntrack_event_cache(IPCT_LABEL, ct);
416 
417 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
418 
419 	return 0;
420 }
421 
422 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
423 			     const struct ovs_key_ct_labels *labels,
424 			     const struct ovs_key_ct_labels *mask)
425 {
426 	struct nf_conn_labels *cl;
427 	int err;
428 
429 	cl = ovs_ct_get_conn_labels(ct);
430 	if (!cl)
431 		return -ENOSPC;
432 
433 	err = nf_connlabels_replace(ct, labels->ct_labels_32,
434 				    mask->ct_labels_32,
435 				    OVS_CT_LABELS_LEN_32);
436 	if (err)
437 		return err;
438 
439 	memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
440 
441 	return 0;
442 }
443 
444 static int ovs_ct_handle_fragments(struct net *net, struct sw_flow_key *key,
445 				   u16 zone, int family, struct sk_buff *skb)
446 {
447 	struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
448 	int err;
449 
450 	err = nf_ct_handle_fragments(net, skb, zone, family, &key->ip.proto, &ovs_cb.mru);
451 	if (err)
452 		return err;
453 
454 	/* The key extracted from the fragment that completed this datagram
455 	 * likely didn't have an L4 header, so regenerate it.
456 	 */
457 	ovs_flow_key_update_l3l4(skb, key);
458 	key->ip.frag = OVS_FRAG_TYPE_NONE;
459 	*OVS_CB(skb) = ovs_cb;
460 
461 	return 0;
462 }
463 
464 /* This replicates logic from nf_conntrack_core.c that is not exported. */
465 static enum ip_conntrack_info
466 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
467 {
468 	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
469 
470 	if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
471 		return IP_CT_ESTABLISHED_REPLY;
472 	/* Once we've had two way comms, always ESTABLISHED. */
473 	if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
474 		return IP_CT_ESTABLISHED;
475 	if (test_bit(IPS_EXPECTED_BIT, &ct->status))
476 		return IP_CT_RELATED;
477 	return IP_CT_NEW;
478 }
479 
480 /* Find an existing connection which this packet belongs to without
481  * re-attributing statistics or modifying the connection state.  This allows an
482  * skb->_nfct lost due to an upcall to be recovered during actions execution.
483  *
484  * Must be called with rcu_read_lock.
485  *
486  * On success, populates skb->_nfct and returns the connection.  Returns NULL
487  * if there is no existing entry.
488  */
489 static struct nf_conn *
490 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
491 		     u8 l3num, struct sk_buff *skb, bool natted)
492 {
493 	struct nf_conntrack_tuple tuple;
494 	struct nf_conntrack_tuple_hash *h;
495 	struct nf_conn *ct;
496 
497 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
498 			       net, &tuple)) {
499 		pr_debug("ovs_ct_find_existing: Can't get tuple\n");
500 		return NULL;
501 	}
502 
503 	/* Must invert the tuple if skb has been transformed by NAT. */
504 	if (natted) {
505 		struct nf_conntrack_tuple inverse;
506 
507 		if (!nf_ct_invert_tuple(&inverse, &tuple)) {
508 			pr_debug("ovs_ct_find_existing: Inversion failed!\n");
509 			return NULL;
510 		}
511 		tuple = inverse;
512 	}
513 
514 	/* look for tuple match */
515 	h = nf_conntrack_find_get(net, zone, &tuple);
516 	if (!h)
517 		return NULL;   /* Not found. */
518 
519 	ct = nf_ct_tuplehash_to_ctrack(h);
520 
521 	/* Inverted packet tuple matches the reverse direction conntrack tuple,
522 	 * select the other tuplehash to get the right 'ctinfo' bits for this
523 	 * packet.
524 	 */
525 	if (natted)
526 		h = &ct->tuplehash[!h->tuple.dst.dir];
527 
528 	nf_ct_set(skb, ct, ovs_ct_get_info(h));
529 	return ct;
530 }
531 
532 static
533 struct nf_conn *ovs_ct_executed(struct net *net,
534 				const struct sw_flow_key *key,
535 				const struct ovs_conntrack_info *info,
536 				struct sk_buff *skb,
537 				bool *ct_executed)
538 {
539 	struct nf_conn *ct = NULL;
540 
541 	/* If no ct, check if we have evidence that an existing conntrack entry
542 	 * might be found for this skb.  This happens when we lose a skb->_nfct
543 	 * due to an upcall, or if the direction is being forced.  If the
544 	 * connection was not confirmed, it is not cached and needs to be run
545 	 * through conntrack again.
546 	 */
547 	*ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
548 		       !(key->ct_state & OVS_CS_F_INVALID) &&
549 		       (key->ct_zone == info->zone.id);
550 
551 	if (*ct_executed || (!key->ct_state && info->force)) {
552 		ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
553 					  !!(key->ct_state &
554 					  OVS_CS_F_NAT_MASK));
555 	}
556 
557 	return ct;
558 }
559 
560 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
561 static bool skb_nfct_cached(struct net *net,
562 			    const struct sw_flow_key *key,
563 			    const struct ovs_conntrack_info *info,
564 			    struct sk_buff *skb)
565 {
566 	enum ip_conntrack_info ctinfo;
567 	struct nf_conn *ct;
568 	bool ct_executed = true;
569 
570 	ct = nf_ct_get(skb, &ctinfo);
571 	if (!ct)
572 		ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
573 
574 	if (ct)
575 		nf_ct_get(skb, &ctinfo);
576 	else
577 		return false;
578 
579 	if (!net_eq(net, read_pnet(&ct->ct_net)))
580 		return false;
581 	if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
582 		return false;
583 	if (info->helper) {
584 		struct nf_conn_help *help;
585 
586 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
587 		if (help && rcu_access_pointer(help->helper) != info->helper)
588 			return false;
589 	}
590 	if (info->nf_ct_timeout) {
591 		struct nf_conn_timeout *timeout_ext;
592 
593 		timeout_ext = nf_ct_timeout_find(ct);
594 		if (!timeout_ext || info->nf_ct_timeout !=
595 		    rcu_dereference(timeout_ext->timeout))
596 			return false;
597 	}
598 	/* Force conntrack entry direction to the current packet? */
599 	if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
600 		/* Delete the conntrack entry if confirmed, else just release
601 		 * the reference.
602 		 */
603 		if (nf_ct_is_confirmed(ct))
604 			nf_ct_delete(ct, 0, 0);
605 
606 		nf_ct_put(ct);
607 		nf_ct_set(skb, NULL, 0);
608 		return false;
609 	}
610 
611 	return ct_executed;
612 }
613 
614 #if IS_ENABLED(CONFIG_NF_NAT)
615 static void ovs_nat_update_key(struct sw_flow_key *key,
616 			       const struct sk_buff *skb,
617 			       enum nf_nat_manip_type maniptype)
618 {
619 	if (maniptype == NF_NAT_MANIP_SRC) {
620 		__be16 src;
621 
622 		key->ct_state |= OVS_CS_F_SRC_NAT;
623 		if (key->eth.type == htons(ETH_P_IP))
624 			key->ipv4.addr.src = ip_hdr(skb)->saddr;
625 		else if (key->eth.type == htons(ETH_P_IPV6))
626 			memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
627 			       sizeof(key->ipv6.addr.src));
628 		else
629 			return;
630 
631 		if (key->ip.proto == IPPROTO_UDP)
632 			src = udp_hdr(skb)->source;
633 		else if (key->ip.proto == IPPROTO_TCP)
634 			src = tcp_hdr(skb)->source;
635 		else if (key->ip.proto == IPPROTO_SCTP)
636 			src = sctp_hdr(skb)->source;
637 		else
638 			return;
639 
640 		key->tp.src = src;
641 	} else {
642 		__be16 dst;
643 
644 		key->ct_state |= OVS_CS_F_DST_NAT;
645 		if (key->eth.type == htons(ETH_P_IP))
646 			key->ipv4.addr.dst = ip_hdr(skb)->daddr;
647 		else if (key->eth.type == htons(ETH_P_IPV6))
648 			memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
649 			       sizeof(key->ipv6.addr.dst));
650 		else
651 			return;
652 
653 		if (key->ip.proto == IPPROTO_UDP)
654 			dst = udp_hdr(skb)->dest;
655 		else if (key->ip.proto == IPPROTO_TCP)
656 			dst = tcp_hdr(skb)->dest;
657 		else if (key->ip.proto == IPPROTO_SCTP)
658 			dst = sctp_hdr(skb)->dest;
659 		else
660 			return;
661 
662 		key->tp.dst = dst;
663 	}
664 }
665 
666 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
667 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
668 		      const struct ovs_conntrack_info *info,
669 		      struct sk_buff *skb, struct nf_conn *ct,
670 		      enum ip_conntrack_info ctinfo)
671 {
672 	int err, action = 0;
673 
674 	if (!(info->nat & OVS_CT_NAT))
675 		return NF_ACCEPT;
676 	if (info->nat & OVS_CT_SRC_NAT)
677 		action |= BIT(NF_NAT_MANIP_SRC);
678 	if (info->nat & OVS_CT_DST_NAT)
679 		action |= BIT(NF_NAT_MANIP_DST);
680 
681 	err = nf_ct_nat(skb, ct, ctinfo, &action, &info->range, info->commit);
682 	if (err != NF_ACCEPT)
683 		return err;
684 
685 	if (action & BIT(NF_NAT_MANIP_SRC))
686 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_SRC);
687 	if (action & BIT(NF_NAT_MANIP_DST))
688 		ovs_nat_update_key(key, skb, NF_NAT_MANIP_DST);
689 
690 	return err;
691 }
692 #else /* !CONFIG_NF_NAT */
693 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
694 		      const struct ovs_conntrack_info *info,
695 		      struct sk_buff *skb, struct nf_conn *ct,
696 		      enum ip_conntrack_info ctinfo)
697 {
698 	return NF_ACCEPT;
699 }
700 #endif
701 
702 static int verdict_to_errno(unsigned int verdict)
703 {
704 	switch (verdict & NF_VERDICT_MASK) {
705 	case NF_ACCEPT:
706 		return 0;
707 	case NF_DROP:
708 		return -EINVAL;
709 	case NF_STOLEN:
710 		return -EINPROGRESS;
711 	default:
712 		break;
713 	}
714 
715 	return -EINVAL;
716 }
717 
718 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
719  * not done already.  Update key with new CT state after passing the packet
720  * through conntrack.
721  * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
722  * set to NULL and 0 will be returned.
723  */
724 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
725 			   const struct ovs_conntrack_info *info,
726 			   struct sk_buff *skb)
727 {
728 	/* If we are recirculating packets to match on conntrack fields and
729 	 * committing with a separate conntrack action,  then we don't need to
730 	 * actually run the packet through conntrack twice unless it's for a
731 	 * different zone.
732 	 */
733 	bool cached = skb_nfct_cached(net, key, info, skb);
734 	enum ip_conntrack_info ctinfo;
735 	struct nf_conn *ct;
736 
737 	if (!cached) {
738 		struct nf_hook_state state = {
739 			.hook = NF_INET_PRE_ROUTING,
740 			.pf = info->family,
741 			.net = net,
742 		};
743 		struct nf_conn *tmpl = info->ct;
744 		int err;
745 
746 		/* Associate skb with specified zone. */
747 		if (tmpl) {
748 			ct = nf_ct_get(skb, &ctinfo);
749 			nf_ct_put(ct);
750 			nf_conntrack_get(&tmpl->ct_general);
751 			nf_ct_set(skb, tmpl, IP_CT_NEW);
752 		}
753 
754 		err = nf_conntrack_in(skb, &state);
755 		if (err != NF_ACCEPT)
756 			return verdict_to_errno(err);
757 
758 		/* Clear CT state NAT flags to mark that we have not yet done
759 		 * NAT after the nf_conntrack_in() call.  We can actually clear
760 		 * the whole state, as it will be re-initialized below.
761 		 */
762 		key->ct_state = 0;
763 
764 		/* Update the key, but keep the NAT flags. */
765 		ovs_ct_update_key(skb, info, key, true, true);
766 	}
767 
768 	ct = nf_ct_get(skb, &ctinfo);
769 	if (ct) {
770 		bool add_helper = false;
771 
772 		/* Packets starting a new connection must be NATted before the
773 		 * helper, so that the helper knows about the NAT.  We enforce
774 		 * this by delaying both NAT and helper calls for unconfirmed
775 		 * connections until the committing CT action.  For later
776 		 * packets NAT and Helper may be called in either order.
777 		 *
778 		 * NAT will be done only if the CT action has NAT, and only
779 		 * once per packet (per zone), as guarded by the NAT bits in
780 		 * the key->ct_state.
781 		 */
782 		if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
783 		    (nf_ct_is_confirmed(ct) || info->commit)) {
784 			int err = ovs_ct_nat(net, key, info, skb, ct, ctinfo);
785 
786 			err = verdict_to_errno(err);
787 			if (err)
788 				return err;
789 		}
790 
791 		/* Userspace may decide to perform a ct lookup without a helper
792 		 * specified followed by a (recirculate and) commit with one,
793 		 * or attach a helper in a later commit.  Therefore, for
794 		 * connections which we will commit, we may need to attach
795 		 * the helper here.
796 		 */
797 		if (!nf_ct_is_confirmed(ct) && info->commit &&
798 		    info->helper && !nfct_help(ct)) {
799 			int err = __nf_ct_try_assign_helper(ct, info->ct,
800 							    GFP_ATOMIC);
801 			if (err)
802 				return err;
803 			add_helper = true;
804 
805 			/* helper installed, add seqadj if NAT is required */
806 			if (info->nat && !nfct_seqadj(ct)) {
807 				if (!nfct_seqadj_ext_add(ct))
808 					return -EINVAL;
809 			}
810 		}
811 
812 		/* Call the helper only if:
813 		 * - nf_conntrack_in() was executed above ("!cached") or a
814 		 *   helper was just attached ("add_helper") for a confirmed
815 		 *   connection, or
816 		 * - When committing an unconfirmed connection.
817 		 */
818 		if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
819 					      info->commit)) {
820 			int err = nf_ct_helper(skb, ct, ctinfo, info->family);
821 
822 			err = verdict_to_errno(err);
823 			if (err)
824 				return err;
825 		}
826 
827 		if (nf_ct_protonum(ct) == IPPROTO_TCP &&
828 		    nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
829 			/* Be liberal for tcp packets so that out-of-window
830 			 * packets are not marked invalid.
831 			 */
832 			nf_ct_set_tcp_be_liberal(ct);
833 		}
834 
835 		nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
836 	}
837 
838 	return 0;
839 }
840 
841 /* Lookup connection and read fields into key. */
842 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
843 			 const struct ovs_conntrack_info *info,
844 			 struct sk_buff *skb)
845 {
846 	struct nf_conn *ct;
847 	int err;
848 
849 	err = __ovs_ct_lookup(net, key, info, skb);
850 	if (err)
851 		return err;
852 
853 	ct = (struct nf_conn *)skb_nfct(skb);
854 	if (ct)
855 		nf_ct_deliver_cached_events(ct);
856 
857 	return 0;
858 }
859 
860 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
861 {
862 	size_t i;
863 
864 	for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
865 		if (labels->ct_labels_32[i])
866 			return true;
867 
868 	return false;
869 }
870 
871 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
872 static struct hlist_head *ct_limit_hash_bucket(
873 	const struct ovs_ct_limit_info *info, u16 zone)
874 {
875 	return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
876 }
877 
878 /* Call with ovs_mutex */
879 static void ct_limit_set(const struct ovs_ct_limit_info *info,
880 			 struct ovs_ct_limit *new_ct_limit)
881 {
882 	struct ovs_ct_limit *ct_limit;
883 	struct hlist_head *head;
884 
885 	head = ct_limit_hash_bucket(info, new_ct_limit->zone);
886 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
887 		if (ct_limit->zone == new_ct_limit->zone) {
888 			hlist_replace_rcu(&ct_limit->hlist_node,
889 					  &new_ct_limit->hlist_node);
890 			kfree_rcu(ct_limit, rcu);
891 			return;
892 		}
893 	}
894 
895 	hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
896 }
897 
898 /* Call with ovs_mutex */
899 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
900 {
901 	struct ovs_ct_limit *ct_limit;
902 	struct hlist_head *head;
903 	struct hlist_node *n;
904 
905 	head = ct_limit_hash_bucket(info, zone);
906 	hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
907 		if (ct_limit->zone == zone) {
908 			hlist_del_rcu(&ct_limit->hlist_node);
909 			kfree_rcu(ct_limit, rcu);
910 			return;
911 		}
912 	}
913 }
914 
915 /* Call with RCU read lock */
916 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
917 {
918 	struct ovs_ct_limit *ct_limit;
919 	struct hlist_head *head;
920 
921 	head = ct_limit_hash_bucket(info, zone);
922 	hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
923 		if (ct_limit->zone == zone)
924 			return ct_limit->limit;
925 	}
926 
927 	return info->default_limit;
928 }
929 
930 static int ovs_ct_check_limit(struct net *net,
931 			      const struct ovs_conntrack_info *info,
932 			      const struct nf_conntrack_tuple *tuple)
933 {
934 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
935 	const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
936 	u32 per_zone_limit, connections;
937 	u32 conncount_key;
938 
939 	conncount_key = info->zone.id;
940 
941 	per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
942 	if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
943 		return 0;
944 
945 	connections = nf_conncount_count(net, ct_limit_info->data,
946 					 &conncount_key, tuple, &info->zone);
947 	if (connections > per_zone_limit)
948 		return -ENOMEM;
949 
950 	return 0;
951 }
952 #endif
953 
954 /* Lookup connection and confirm if unconfirmed. */
955 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
956 			 const struct ovs_conntrack_info *info,
957 			 struct sk_buff *skb)
958 {
959 	enum ip_conntrack_info ctinfo;
960 	struct nf_conn *ct;
961 	int err;
962 
963 	err = __ovs_ct_lookup(net, key, info, skb);
964 	if (err)
965 		return err;
966 
967 	/* The connection could be invalid, in which case this is a no-op.*/
968 	ct = nf_ct_get(skb, &ctinfo);
969 	if (!ct)
970 		return 0;
971 
972 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
973 	if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
974 		if (!nf_ct_is_confirmed(ct)) {
975 			err = ovs_ct_check_limit(net, info,
976 				&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
977 			if (err) {
978 				net_warn_ratelimited("openvswitch: zone: %u "
979 					"exceeds conntrack limit\n",
980 					info->zone.id);
981 				return err;
982 			}
983 		}
984 	}
985 #endif
986 
987 	/* Set the conntrack event mask if given.  NEW and DELETE events have
988 	 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
989 	 * typically would receive many kinds of updates.  Setting the event
990 	 * mask allows those events to be filtered.  The set event mask will
991 	 * remain in effect for the lifetime of the connection unless changed
992 	 * by a further CT action with both the commit flag and the eventmask
993 	 * option. */
994 	if (info->have_eventmask) {
995 		struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
996 
997 		if (cache)
998 			cache->ctmask = info->eventmask;
999 	}
1000 
1001 	/* Apply changes before confirming the connection so that the initial
1002 	 * conntrack NEW netlink event carries the values given in the CT
1003 	 * action.
1004 	 */
1005 	if (info->mark.mask) {
1006 		err = ovs_ct_set_mark(ct, key, info->mark.value,
1007 				      info->mark.mask);
1008 		if (err)
1009 			return err;
1010 	}
1011 	if (!nf_ct_is_confirmed(ct)) {
1012 		err = ovs_ct_init_labels(ct, key, &info->labels.value,
1013 					 &info->labels.mask);
1014 		if (err)
1015 			return err;
1016 
1017 		nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1018 	} else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1019 		   labels_nonzero(&info->labels.mask)) {
1020 		err = ovs_ct_set_labels(ct, key, &info->labels.value,
1021 					&info->labels.mask);
1022 		if (err)
1023 			return err;
1024 	}
1025 	/* This will take care of sending queued events even if the connection
1026 	 * is already confirmed.
1027 	 */
1028 	err = nf_conntrack_confirm(skb);
1029 
1030 	return verdict_to_errno(err);
1031 }
1032 
1033 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1034  * value if 'skb' is freed.
1035  */
1036 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1037 		   struct sw_flow_key *key,
1038 		   const struct ovs_conntrack_info *info)
1039 {
1040 	int nh_ofs;
1041 	int err;
1042 
1043 	/* The conntrack module expects to be working at L3. */
1044 	nh_ofs = skb_network_offset(skb);
1045 	skb_pull_rcsum(skb, nh_ofs);
1046 
1047 	err = nf_ct_skb_network_trim(skb, info->family);
1048 	if (err) {
1049 		kfree_skb(skb);
1050 		return err;
1051 	}
1052 
1053 	if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1054 		err = ovs_ct_handle_fragments(net, key, info->zone.id,
1055 					      info->family, skb);
1056 		if (err)
1057 			return err;
1058 	}
1059 
1060 	if (info->commit)
1061 		err = ovs_ct_commit(net, key, info, skb);
1062 	else
1063 		err = ovs_ct_lookup(net, key, info, skb);
1064 
1065 	/* conntrack core returned NF_STOLEN */
1066 	if (err == -EINPROGRESS)
1067 		return err;
1068 
1069 	skb_push_rcsum(skb, nh_ofs);
1070 	if (err)
1071 		ovs_kfree_skb_reason(skb, OVS_DROP_CONNTRACK);
1072 	return err;
1073 }
1074 
1075 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1076 {
1077 	enum ip_conntrack_info ctinfo;
1078 	struct nf_conn *ct;
1079 
1080 	ct = nf_ct_get(skb, &ctinfo);
1081 
1082 	nf_ct_put(ct);
1083 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1084 
1085 	if (key)
1086 		ovs_ct_fill_key(skb, key, false);
1087 
1088 	return 0;
1089 }
1090 
1091 #if IS_ENABLED(CONFIG_NF_NAT)
1092 static int parse_nat(const struct nlattr *attr,
1093 		     struct ovs_conntrack_info *info, bool log)
1094 {
1095 	struct nlattr *a;
1096 	int rem;
1097 	bool have_ip_max = false;
1098 	bool have_proto_max = false;
1099 	bool ip_vers = (info->family == NFPROTO_IPV6);
1100 
1101 	nla_for_each_nested(a, attr, rem) {
1102 		static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1103 			[OVS_NAT_ATTR_SRC] = {0, 0},
1104 			[OVS_NAT_ATTR_DST] = {0, 0},
1105 			[OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1106 						 sizeof(struct in6_addr)},
1107 			[OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1108 						 sizeof(struct in6_addr)},
1109 			[OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1110 			[OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1111 			[OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1112 			[OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1113 			[OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1114 		};
1115 		int type = nla_type(a);
1116 
1117 		if (type > OVS_NAT_ATTR_MAX) {
1118 			OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1119 				  type, OVS_NAT_ATTR_MAX);
1120 			return -EINVAL;
1121 		}
1122 
1123 		if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1124 			OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1125 				  type, nla_len(a),
1126 				  ovs_nat_attr_lens[type][ip_vers]);
1127 			return -EINVAL;
1128 		}
1129 
1130 		switch (type) {
1131 		case OVS_NAT_ATTR_SRC:
1132 		case OVS_NAT_ATTR_DST:
1133 			if (info->nat) {
1134 				OVS_NLERR(log, "Only one type of NAT may be specified");
1135 				return -ERANGE;
1136 			}
1137 			info->nat |= OVS_CT_NAT;
1138 			info->nat |= ((type == OVS_NAT_ATTR_SRC)
1139 					? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1140 			break;
1141 
1142 		case OVS_NAT_ATTR_IP_MIN:
1143 			nla_memcpy(&info->range.min_addr, a,
1144 				   sizeof(info->range.min_addr));
1145 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1146 			break;
1147 
1148 		case OVS_NAT_ATTR_IP_MAX:
1149 			have_ip_max = true;
1150 			nla_memcpy(&info->range.max_addr, a,
1151 				   sizeof(info->range.max_addr));
1152 			info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1153 			break;
1154 
1155 		case OVS_NAT_ATTR_PROTO_MIN:
1156 			info->range.min_proto.all = htons(nla_get_u16(a));
1157 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1158 			break;
1159 
1160 		case OVS_NAT_ATTR_PROTO_MAX:
1161 			have_proto_max = true;
1162 			info->range.max_proto.all = htons(nla_get_u16(a));
1163 			info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1164 			break;
1165 
1166 		case OVS_NAT_ATTR_PERSISTENT:
1167 			info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1168 			break;
1169 
1170 		case OVS_NAT_ATTR_PROTO_HASH:
1171 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1172 			break;
1173 
1174 		case OVS_NAT_ATTR_PROTO_RANDOM:
1175 			info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1176 			break;
1177 
1178 		default:
1179 			OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1180 			return -EINVAL;
1181 		}
1182 	}
1183 
1184 	if (rem > 0) {
1185 		OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1186 		return -EINVAL;
1187 	}
1188 	if (!info->nat) {
1189 		/* Do not allow flags if no type is given. */
1190 		if (info->range.flags) {
1191 			OVS_NLERR(log,
1192 				  "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1193 				  );
1194 			return -EINVAL;
1195 		}
1196 		info->nat = OVS_CT_NAT;   /* NAT existing connections. */
1197 	} else if (!info->commit) {
1198 		OVS_NLERR(log,
1199 			  "NAT attributes may be specified only when CT COMMIT flag is also specified."
1200 			  );
1201 		return -EINVAL;
1202 	}
1203 	/* Allow missing IP_MAX. */
1204 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1205 		memcpy(&info->range.max_addr, &info->range.min_addr,
1206 		       sizeof(info->range.max_addr));
1207 	}
1208 	/* Allow missing PROTO_MAX. */
1209 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1210 	    !have_proto_max) {
1211 		info->range.max_proto.all = info->range.min_proto.all;
1212 	}
1213 	return 0;
1214 }
1215 #endif
1216 
1217 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1218 	[OVS_CT_ATTR_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1219 	[OVS_CT_ATTR_FORCE_COMMIT]	= { .minlen = 0, .maxlen = 0 },
1220 	[OVS_CT_ATTR_ZONE]	= { .minlen = sizeof(u16),
1221 				    .maxlen = sizeof(u16) },
1222 	[OVS_CT_ATTR_MARK]	= { .minlen = sizeof(struct md_mark),
1223 				    .maxlen = sizeof(struct md_mark) },
1224 	[OVS_CT_ATTR_LABELS]	= { .minlen = sizeof(struct md_labels),
1225 				    .maxlen = sizeof(struct md_labels) },
1226 	[OVS_CT_ATTR_HELPER]	= { .minlen = 1,
1227 				    .maxlen = NF_CT_HELPER_NAME_LEN },
1228 #if IS_ENABLED(CONFIG_NF_NAT)
1229 	/* NAT length is checked when parsing the nested attributes. */
1230 	[OVS_CT_ATTR_NAT]	= { .minlen = 0, .maxlen = INT_MAX },
1231 #endif
1232 	[OVS_CT_ATTR_EVENTMASK]	= { .minlen = sizeof(u32),
1233 				    .maxlen = sizeof(u32) },
1234 	[OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1235 				  .maxlen = CTNL_TIMEOUT_NAME_MAX },
1236 };
1237 
1238 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1239 		    const char **helper, bool log)
1240 {
1241 	struct nlattr *a;
1242 	int rem;
1243 
1244 	nla_for_each_nested(a, attr, rem) {
1245 		int type = nla_type(a);
1246 		int maxlen;
1247 		int minlen;
1248 
1249 		if (type > OVS_CT_ATTR_MAX) {
1250 			OVS_NLERR(log,
1251 				  "Unknown conntrack attr (type=%d, max=%d)",
1252 				  type, OVS_CT_ATTR_MAX);
1253 			return -EINVAL;
1254 		}
1255 
1256 		maxlen = ovs_ct_attr_lens[type].maxlen;
1257 		minlen = ovs_ct_attr_lens[type].minlen;
1258 		if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1259 			OVS_NLERR(log,
1260 				  "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1261 				  type, nla_len(a), maxlen);
1262 			return -EINVAL;
1263 		}
1264 
1265 		switch (type) {
1266 		case OVS_CT_ATTR_FORCE_COMMIT:
1267 			info->force = true;
1268 			fallthrough;
1269 		case OVS_CT_ATTR_COMMIT:
1270 			info->commit = true;
1271 			break;
1272 #ifdef CONFIG_NF_CONNTRACK_ZONES
1273 		case OVS_CT_ATTR_ZONE:
1274 			info->zone.id = nla_get_u16(a);
1275 			break;
1276 #endif
1277 #ifdef CONFIG_NF_CONNTRACK_MARK
1278 		case OVS_CT_ATTR_MARK: {
1279 			struct md_mark *mark = nla_data(a);
1280 
1281 			if (!mark->mask) {
1282 				OVS_NLERR(log, "ct_mark mask cannot be 0");
1283 				return -EINVAL;
1284 			}
1285 			info->mark = *mark;
1286 			break;
1287 		}
1288 #endif
1289 #ifdef CONFIG_NF_CONNTRACK_LABELS
1290 		case OVS_CT_ATTR_LABELS: {
1291 			struct md_labels *labels = nla_data(a);
1292 
1293 			if (!labels_nonzero(&labels->mask)) {
1294 				OVS_NLERR(log, "ct_labels mask cannot be 0");
1295 				return -EINVAL;
1296 			}
1297 			info->labels = *labels;
1298 			break;
1299 		}
1300 #endif
1301 		case OVS_CT_ATTR_HELPER:
1302 			*helper = nla_data(a);
1303 			if (!string_is_terminated(*helper, nla_len(a))) {
1304 				OVS_NLERR(log, "Invalid conntrack helper");
1305 				return -EINVAL;
1306 			}
1307 			break;
1308 #if IS_ENABLED(CONFIG_NF_NAT)
1309 		case OVS_CT_ATTR_NAT: {
1310 			int err = parse_nat(a, info, log);
1311 
1312 			if (err)
1313 				return err;
1314 			break;
1315 		}
1316 #endif
1317 		case OVS_CT_ATTR_EVENTMASK:
1318 			info->have_eventmask = true;
1319 			info->eventmask = nla_get_u32(a);
1320 			break;
1321 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1322 		case OVS_CT_ATTR_TIMEOUT:
1323 			memcpy(info->timeout, nla_data(a), nla_len(a));
1324 			if (!string_is_terminated(info->timeout, nla_len(a))) {
1325 				OVS_NLERR(log, "Invalid conntrack timeout");
1326 				return -EINVAL;
1327 			}
1328 			break;
1329 #endif
1330 
1331 		default:
1332 			OVS_NLERR(log, "Unknown conntrack attr (%d)",
1333 				  type);
1334 			return -EINVAL;
1335 		}
1336 	}
1337 
1338 #ifdef CONFIG_NF_CONNTRACK_MARK
1339 	if (!info->commit && info->mark.mask) {
1340 		OVS_NLERR(log,
1341 			  "Setting conntrack mark requires 'commit' flag.");
1342 		return -EINVAL;
1343 	}
1344 #endif
1345 #ifdef CONFIG_NF_CONNTRACK_LABELS
1346 	if (!info->commit && labels_nonzero(&info->labels.mask)) {
1347 		OVS_NLERR(log,
1348 			  "Setting conntrack labels requires 'commit' flag.");
1349 		return -EINVAL;
1350 	}
1351 #endif
1352 	if (rem > 0) {
1353 		OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1354 		return -EINVAL;
1355 	}
1356 
1357 	return 0;
1358 }
1359 
1360 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1361 {
1362 	if (attr == OVS_KEY_ATTR_CT_STATE)
1363 		return true;
1364 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1365 	    attr == OVS_KEY_ATTR_CT_ZONE)
1366 		return true;
1367 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1368 	    attr == OVS_KEY_ATTR_CT_MARK)
1369 		return true;
1370 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1371 	    attr == OVS_KEY_ATTR_CT_LABELS)
1372 		return true;
1373 
1374 	return false;
1375 }
1376 
1377 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1378 		       const struct sw_flow_key *key,
1379 		       struct sw_flow_actions **sfa,  bool log)
1380 {
1381 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1382 	struct ovs_conntrack_info ct_info;
1383 	const char *helper = NULL;
1384 	u16 family;
1385 	int err;
1386 
1387 	family = key_to_nfproto(key);
1388 	if (family == NFPROTO_UNSPEC) {
1389 		OVS_NLERR(log, "ct family unspecified");
1390 		return -EINVAL;
1391 	}
1392 
1393 	memset(&ct_info, 0, sizeof(ct_info));
1394 	ct_info.family = family;
1395 
1396 	nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1397 			NF_CT_DEFAULT_ZONE_DIR, 0);
1398 
1399 	err = parse_ct(attr, &ct_info, &helper, log);
1400 	if (err)
1401 		return err;
1402 
1403 	/* Set up template for tracking connections in specific zones. */
1404 	ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1405 	if (!ct_info.ct) {
1406 		OVS_NLERR(log, "Failed to allocate conntrack template");
1407 		return -ENOMEM;
1408 	}
1409 
1410 	if (nf_connlabels_get(net, n_bits - 1)) {
1411 		nf_ct_tmpl_free(ct_info.ct);
1412 		OVS_NLERR(log, "Failed to set connlabel length");
1413 		return -EOPNOTSUPP;
1414 	}
1415 
1416 	if (ct_info.timeout[0]) {
1417 		if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1418 				      ct_info.timeout))
1419 			OVS_NLERR(log,
1420 				  "Failed to associated timeout policy '%s'",
1421 				  ct_info.timeout);
1422 		else
1423 			ct_info.nf_ct_timeout = rcu_dereference(
1424 				nf_ct_timeout_find(ct_info.ct)->timeout);
1425 
1426 	}
1427 
1428 	if (helper) {
1429 		err = nf_ct_add_helper(ct_info.ct, helper, ct_info.family,
1430 				       key->ip.proto, ct_info.nat, &ct_info.helper);
1431 		if (err) {
1432 			OVS_NLERR(log, "Failed to add %s helper %d", helper, err);
1433 			goto err_free_ct;
1434 		}
1435 	}
1436 
1437 	err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1438 				 sizeof(ct_info), log);
1439 	if (err)
1440 		goto err_free_ct;
1441 
1442 	if (ct_info.commit)
1443 		__set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1444 	return 0;
1445 err_free_ct:
1446 	__ovs_ct_free_action(&ct_info);
1447 	return err;
1448 }
1449 
1450 #if IS_ENABLED(CONFIG_NF_NAT)
1451 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1452 			       struct sk_buff *skb)
1453 {
1454 	struct nlattr *start;
1455 
1456 	start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1457 	if (!start)
1458 		return false;
1459 
1460 	if (info->nat & OVS_CT_SRC_NAT) {
1461 		if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1462 			return false;
1463 	} else if (info->nat & OVS_CT_DST_NAT) {
1464 		if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1465 			return false;
1466 	} else {
1467 		goto out;
1468 	}
1469 
1470 	if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1471 		if (IS_ENABLED(CONFIG_NF_NAT) &&
1472 		    info->family == NFPROTO_IPV4) {
1473 			if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1474 					    info->range.min_addr.ip) ||
1475 			    (info->range.max_addr.ip
1476 			     != info->range.min_addr.ip &&
1477 			     (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1478 					      info->range.max_addr.ip))))
1479 				return false;
1480 		} else if (IS_ENABLED(CONFIG_IPV6) &&
1481 			   info->family == NFPROTO_IPV6) {
1482 			if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1483 					     &info->range.min_addr.in6) ||
1484 			    (memcmp(&info->range.max_addr.in6,
1485 				    &info->range.min_addr.in6,
1486 				    sizeof(info->range.max_addr.in6)) &&
1487 			     (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1488 					       &info->range.max_addr.in6))))
1489 				return false;
1490 		} else {
1491 			return false;
1492 		}
1493 	}
1494 	if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1495 	    (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1496 			 ntohs(info->range.min_proto.all)) ||
1497 	     (info->range.max_proto.all != info->range.min_proto.all &&
1498 	      nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1499 			  ntohs(info->range.max_proto.all)))))
1500 		return false;
1501 
1502 	if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1503 	    nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1504 		return false;
1505 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1506 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1507 		return false;
1508 	if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1509 	    nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1510 		return false;
1511 out:
1512 	nla_nest_end(skb, start);
1513 
1514 	return true;
1515 }
1516 #endif
1517 
1518 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1519 			  struct sk_buff *skb)
1520 {
1521 	struct nlattr *start;
1522 
1523 	start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1524 	if (!start)
1525 		return -EMSGSIZE;
1526 
1527 	if (ct_info->commit && nla_put_flag(skb, ct_info->force
1528 					    ? OVS_CT_ATTR_FORCE_COMMIT
1529 					    : OVS_CT_ATTR_COMMIT))
1530 		return -EMSGSIZE;
1531 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1532 	    nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1533 		return -EMSGSIZE;
1534 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1535 	    nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1536 		    &ct_info->mark))
1537 		return -EMSGSIZE;
1538 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1539 	    labels_nonzero(&ct_info->labels.mask) &&
1540 	    nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1541 		    &ct_info->labels))
1542 		return -EMSGSIZE;
1543 	if (ct_info->helper) {
1544 		if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1545 				   ct_info->helper->name))
1546 			return -EMSGSIZE;
1547 	}
1548 	if (ct_info->have_eventmask &&
1549 	    nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1550 		return -EMSGSIZE;
1551 	if (ct_info->timeout[0]) {
1552 		if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1553 			return -EMSGSIZE;
1554 	}
1555 
1556 #if IS_ENABLED(CONFIG_NF_NAT)
1557 	if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1558 		return -EMSGSIZE;
1559 #endif
1560 	nla_nest_end(skb, start);
1561 
1562 	return 0;
1563 }
1564 
1565 void ovs_ct_free_action(const struct nlattr *a)
1566 {
1567 	struct ovs_conntrack_info *ct_info = nla_data(a);
1568 
1569 	__ovs_ct_free_action(ct_info);
1570 }
1571 
1572 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1573 {
1574 	if (ct_info->helper) {
1575 #if IS_ENABLED(CONFIG_NF_NAT)
1576 		if (ct_info->nat)
1577 			nf_nat_helper_put(ct_info->helper);
1578 #endif
1579 		nf_conntrack_helper_put(ct_info->helper);
1580 	}
1581 	if (ct_info->ct) {
1582 		if (ct_info->timeout[0])
1583 			nf_ct_destroy_timeout(ct_info->ct);
1584 		nf_connlabels_put(nf_ct_net(ct_info->ct));
1585 		nf_ct_tmpl_free(ct_info->ct);
1586 	}
1587 }
1588 
1589 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1590 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1591 {
1592 	int i, err;
1593 
1594 	ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1595 					 GFP_KERNEL);
1596 	if (!ovs_net->ct_limit_info)
1597 		return -ENOMEM;
1598 
1599 	ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1600 	ovs_net->ct_limit_info->limits =
1601 		kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1602 			      GFP_KERNEL);
1603 	if (!ovs_net->ct_limit_info->limits) {
1604 		kfree(ovs_net->ct_limit_info);
1605 		return -ENOMEM;
1606 	}
1607 
1608 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1609 		INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1610 
1611 	ovs_net->ct_limit_info->data = nf_conncount_init(net, sizeof(u32));
1612 
1613 	if (IS_ERR(ovs_net->ct_limit_info->data)) {
1614 		err = PTR_ERR(ovs_net->ct_limit_info->data);
1615 		kfree(ovs_net->ct_limit_info->limits);
1616 		kfree(ovs_net->ct_limit_info);
1617 		pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1618 		return err;
1619 	}
1620 	return 0;
1621 }
1622 
1623 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1624 {
1625 	const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1626 	int i;
1627 
1628 	nf_conncount_destroy(net, info->data);
1629 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1630 		struct hlist_head *head = &info->limits[i];
1631 		struct ovs_ct_limit *ct_limit;
1632 		struct hlist_node *next;
1633 
1634 		hlist_for_each_entry_safe(ct_limit, next, head, hlist_node)
1635 			kfree_rcu(ct_limit, rcu);
1636 	}
1637 	kfree(info->limits);
1638 	kfree(info);
1639 }
1640 
1641 static struct sk_buff *
1642 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1643 			     struct ovs_header **ovs_reply_header)
1644 {
1645 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1646 	struct sk_buff *skb;
1647 
1648 	skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1649 	if (!skb)
1650 		return ERR_PTR(-ENOMEM);
1651 
1652 	*ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1653 					info->snd_seq,
1654 					&dp_ct_limit_genl_family, 0, cmd);
1655 
1656 	if (!*ovs_reply_header) {
1657 		nlmsg_free(skb);
1658 		return ERR_PTR(-EMSGSIZE);
1659 	}
1660 	(*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1661 
1662 	return skb;
1663 }
1664 
1665 static bool check_zone_id(int zone_id, u16 *pzone)
1666 {
1667 	if (zone_id >= 0 && zone_id <= 65535) {
1668 		*pzone = (u16)zone_id;
1669 		return true;
1670 	}
1671 	return false;
1672 }
1673 
1674 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1675 				       struct ovs_ct_limit_info *info)
1676 {
1677 	struct ovs_zone_limit *zone_limit;
1678 	int rem;
1679 	u16 zone;
1680 
1681 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1682 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1683 
1684 	while (rem >= sizeof(*zone_limit)) {
1685 		if (unlikely(zone_limit->zone_id ==
1686 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1687 			ovs_lock();
1688 			info->default_limit = zone_limit->limit;
1689 			ovs_unlock();
1690 		} else if (unlikely(!check_zone_id(
1691 				zone_limit->zone_id, &zone))) {
1692 			OVS_NLERR(true, "zone id is out of range");
1693 		} else {
1694 			struct ovs_ct_limit *ct_limit;
1695 
1696 			ct_limit = kmalloc(sizeof(*ct_limit),
1697 					   GFP_KERNEL_ACCOUNT);
1698 			if (!ct_limit)
1699 				return -ENOMEM;
1700 
1701 			ct_limit->zone = zone;
1702 			ct_limit->limit = zone_limit->limit;
1703 
1704 			ovs_lock();
1705 			ct_limit_set(info, ct_limit);
1706 			ovs_unlock();
1707 		}
1708 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1709 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1710 				NLA_ALIGN(sizeof(*zone_limit)));
1711 	}
1712 
1713 	if (rem)
1714 		OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1715 
1716 	return 0;
1717 }
1718 
1719 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1720 				       struct ovs_ct_limit_info *info)
1721 {
1722 	struct ovs_zone_limit *zone_limit;
1723 	int rem;
1724 	u16 zone;
1725 
1726 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1727 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1728 
1729 	while (rem >= sizeof(*zone_limit)) {
1730 		if (unlikely(zone_limit->zone_id ==
1731 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1732 			ovs_lock();
1733 			info->default_limit = OVS_CT_LIMIT_DEFAULT;
1734 			ovs_unlock();
1735 		} else if (unlikely(!check_zone_id(
1736 				zone_limit->zone_id, &zone))) {
1737 			OVS_NLERR(true, "zone id is out of range");
1738 		} else {
1739 			ovs_lock();
1740 			ct_limit_del(info, zone);
1741 			ovs_unlock();
1742 		}
1743 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1744 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1745 				NLA_ALIGN(sizeof(*zone_limit)));
1746 	}
1747 
1748 	if (rem)
1749 		OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1750 
1751 	return 0;
1752 }
1753 
1754 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1755 					  struct sk_buff *reply)
1756 {
1757 	struct ovs_zone_limit zone_limit = {
1758 		.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
1759 		.limit   = info->default_limit,
1760 	};
1761 
1762 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1763 }
1764 
1765 static int __ovs_ct_limit_get_zone_limit(struct net *net,
1766 					 struct nf_conncount_data *data,
1767 					 u16 zone_id, u32 limit,
1768 					 struct sk_buff *reply)
1769 {
1770 	struct nf_conntrack_zone ct_zone;
1771 	struct ovs_zone_limit zone_limit;
1772 	u32 conncount_key = zone_id;
1773 
1774 	zone_limit.zone_id = zone_id;
1775 	zone_limit.limit = limit;
1776 	nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1777 
1778 	zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
1779 					      &ct_zone);
1780 	return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1781 }
1782 
1783 static int ovs_ct_limit_get_zone_limit(struct net *net,
1784 				       struct nlattr *nla_zone_limit,
1785 				       struct ovs_ct_limit_info *info,
1786 				       struct sk_buff *reply)
1787 {
1788 	struct ovs_zone_limit *zone_limit;
1789 	int rem, err;
1790 	u32 limit;
1791 	u16 zone;
1792 
1793 	rem = NLA_ALIGN(nla_len(nla_zone_limit));
1794 	zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1795 
1796 	while (rem >= sizeof(*zone_limit)) {
1797 		if (unlikely(zone_limit->zone_id ==
1798 				OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1799 			err = ovs_ct_limit_get_default_limit(info, reply);
1800 			if (err)
1801 				return err;
1802 		} else if (unlikely(!check_zone_id(zone_limit->zone_id,
1803 							&zone))) {
1804 			OVS_NLERR(true, "zone id is out of range");
1805 		} else {
1806 			rcu_read_lock();
1807 			limit = ct_limit_get(info, zone);
1808 			rcu_read_unlock();
1809 
1810 			err = __ovs_ct_limit_get_zone_limit(
1811 				net, info->data, zone, limit, reply);
1812 			if (err)
1813 				return err;
1814 		}
1815 		rem -= NLA_ALIGN(sizeof(*zone_limit));
1816 		zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1817 				NLA_ALIGN(sizeof(*zone_limit)));
1818 	}
1819 
1820 	if (rem)
1821 		OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
1822 
1823 	return 0;
1824 }
1825 
1826 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
1827 					   struct ovs_ct_limit_info *info,
1828 					   struct sk_buff *reply)
1829 {
1830 	struct ovs_ct_limit *ct_limit;
1831 	struct hlist_head *head;
1832 	int i, err = 0;
1833 
1834 	err = ovs_ct_limit_get_default_limit(info, reply);
1835 	if (err)
1836 		return err;
1837 
1838 	rcu_read_lock();
1839 	for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1840 		head = &info->limits[i];
1841 		hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1842 			err = __ovs_ct_limit_get_zone_limit(net, info->data,
1843 				ct_limit->zone, ct_limit->limit, reply);
1844 			if (err)
1845 				goto exit_err;
1846 		}
1847 	}
1848 
1849 exit_err:
1850 	rcu_read_unlock();
1851 	return err;
1852 }
1853 
1854 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
1855 {
1856 	struct nlattr **a = info->attrs;
1857 	struct sk_buff *reply;
1858 	struct ovs_header *ovs_reply_header;
1859 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1860 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1861 	int err;
1862 
1863 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
1864 					     &ovs_reply_header);
1865 	if (IS_ERR(reply))
1866 		return PTR_ERR(reply);
1867 
1868 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1869 		err = -EINVAL;
1870 		goto exit_err;
1871 	}
1872 
1873 	err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1874 					  ct_limit_info);
1875 	if (err)
1876 		goto exit_err;
1877 
1878 	static_branch_enable(&ovs_ct_limit_enabled);
1879 
1880 	genlmsg_end(reply, ovs_reply_header);
1881 	return genlmsg_reply(reply, info);
1882 
1883 exit_err:
1884 	nlmsg_free(reply);
1885 	return err;
1886 }
1887 
1888 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
1889 {
1890 	struct nlattr **a = info->attrs;
1891 	struct sk_buff *reply;
1892 	struct ovs_header *ovs_reply_header;
1893 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1894 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1895 	int err;
1896 
1897 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
1898 					     &ovs_reply_header);
1899 	if (IS_ERR(reply))
1900 		return PTR_ERR(reply);
1901 
1902 	if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1903 		err = -EINVAL;
1904 		goto exit_err;
1905 	}
1906 
1907 	err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
1908 					  ct_limit_info);
1909 	if (err)
1910 		goto exit_err;
1911 
1912 	genlmsg_end(reply, ovs_reply_header);
1913 	return genlmsg_reply(reply, info);
1914 
1915 exit_err:
1916 	nlmsg_free(reply);
1917 	return err;
1918 }
1919 
1920 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
1921 {
1922 	struct nlattr **a = info->attrs;
1923 	struct nlattr *nla_reply;
1924 	struct sk_buff *reply;
1925 	struct ovs_header *ovs_reply_header;
1926 	struct net *net = sock_net(skb->sk);
1927 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1928 	struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1929 	int err;
1930 
1931 	reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
1932 					     &ovs_reply_header);
1933 	if (IS_ERR(reply))
1934 		return PTR_ERR(reply);
1935 
1936 	nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
1937 	if (!nla_reply) {
1938 		err = -EMSGSIZE;
1939 		goto exit_err;
1940 	}
1941 
1942 	if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1943 		err = ovs_ct_limit_get_zone_limit(
1944 			net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
1945 			reply);
1946 		if (err)
1947 			goto exit_err;
1948 	} else {
1949 		err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
1950 						      reply);
1951 		if (err)
1952 			goto exit_err;
1953 	}
1954 
1955 	nla_nest_end(reply, nla_reply);
1956 	genlmsg_end(reply, ovs_reply_header);
1957 	return genlmsg_reply(reply, info);
1958 
1959 exit_err:
1960 	nlmsg_free(reply);
1961 	return err;
1962 }
1963 
1964 static const struct genl_small_ops ct_limit_genl_ops[] = {
1965 	{ .cmd = OVS_CT_LIMIT_CMD_SET,
1966 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1967 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1968 					       * privilege.
1969 					       */
1970 		.doit = ovs_ct_limit_cmd_set,
1971 	},
1972 	{ .cmd = OVS_CT_LIMIT_CMD_DEL,
1973 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1974 		.flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1975 					       * privilege.
1976 					       */
1977 		.doit = ovs_ct_limit_cmd_del,
1978 	},
1979 	{ .cmd = OVS_CT_LIMIT_CMD_GET,
1980 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1981 		.flags = 0,		  /* OK for unprivileged users. */
1982 		.doit = ovs_ct_limit_cmd_get,
1983 	},
1984 };
1985 
1986 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
1987 	.name = OVS_CT_LIMIT_MCGROUP,
1988 };
1989 
1990 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
1991 	.hdrsize = sizeof(struct ovs_header),
1992 	.name = OVS_CT_LIMIT_FAMILY,
1993 	.version = OVS_CT_LIMIT_VERSION,
1994 	.maxattr = OVS_CT_LIMIT_ATTR_MAX,
1995 	.policy = ct_limit_policy,
1996 	.netnsok = true,
1997 	.parallel_ops = true,
1998 	.small_ops = ct_limit_genl_ops,
1999 	.n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2000 	.resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
2001 	.mcgrps = &ovs_ct_limit_multicast_group,
2002 	.n_mcgrps = 1,
2003 	.module = THIS_MODULE,
2004 };
2005 #endif
2006 
2007 int ovs_ct_init(struct net *net)
2008 {
2009 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2010 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2011 
2012 	return ovs_ct_limit_init(net, ovs_net);
2013 #else
2014 	return 0;
2015 #endif
2016 }
2017 
2018 void ovs_ct_exit(struct net *net)
2019 {
2020 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2021 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2022 
2023 	ovs_ct_limit_exit(net, ovs_net);
2024 #endif
2025 }
2026